CoChat

Deep reinforcement learning-based adaptive fuzzy control for electro-hydraulic servo system

Open access
Khater, A. Aziz, Fekry, Mohamed, El-Bardini, Mohammad, El-Nagar, Ahmad M. · 2025 · Neural Computing and Applications

The DOI matches its Crossref record; 42 of the 42 references it cites were found in a registry or an index; 1 of the 26 sources its claims cite were checked for support.

Abstract

In this paper, a novel adaptive fuzzy controller based on deep reinforcement learning (DRL) is introduced for electro-hydraulic servo systems. The controller combines the strengths of fuzzy proportional–integral (PI) control and deep Q-learning network (DQLN) to achieve real-time adaptation and improve the control performance. The purpose of this paper is to address the challenges of controlling electro-hydraulic servo systems by developing an adaptive controller that can dynamically adjust its control parameters based on the system’s state. The traditional fuzzy PI controller is enhanced with DRL techniques to enable automatic adaptation and compensation for changing online conditions. The proposed adaptive controller utilizes a DQLN to dynamically adjust the scaling factors of the input/output membership functions. By using the DQLN algorithm, the controller learns from a variety of system data to determine the optimal control parameters. The update equation of the weights for the Q-network is derived using the Lyapunov stability (LS) theorem, which overcomes the limitations of gradient descent (GD) methods such as instability and local minima trapping. To evaluate the effectiveness of the proposed controller, it is practically implemented to regulate an electro-hydraulic servo system. The controller’s performance is compared against other existing controllers, and its enhancements are demonstrated through experimental evaluation.

Overview

CoChat Summary

TL;DR

The paper proposes AFPIC-DQLN, an adaptive fuzzy PI controller whose input/output scaling factors are tuned online by a deep Q-learning network with Lyapunov-stability-based weight updates, and implements it on an EHS160 electro-hydraulic servo system. In practical tests, it achieved lower IAE and ISE than PID, FPI, AFPIC-QL, and DACRL controllers across step, load, disturbance, and noise tasks, at a higher computation time (0.651 ms) that remained below the sampling period.

Key findings

  • Across all four practical tasks, AFPIC-DQLN had the lowest IAE among the five controllers; e.g., Task 1 IAE was 180.50 vs PID 1788.43, FPI 1280.60, AFPIC-QL 750.33, and DACRL 200.50, and Task 4 IAE was 34.45 vs PID 250.74, FPI 155.12, AFPIC-QL 90.27, and DACRL 40.89.
  • AFPIC-DQLN also had the lowest ISE; e.g., Task 1 ISE was 908.92 vs PID 6100, FPI 5550, AFPIC-QL 2312, and DACRL 1002, and Task 4 ISE was 70.21 vs PID 450.24, FPI 351.36, AFPIC-QL 150.34, and DACRL 112.24.
  • The proposed controller’s computation time was 0.651 ms, higher than PID (0.045 ms), FPI (0.058 ms), AFPIC-QL (0.554 ms), and DACRL (0.597 ms), but the authors state it remains smaller than the sampling period.
  • The Lyapunov-stability-based weight update is presented as overcoming gradient-descent limitations such as instability and local minima, and the controller is reported to adapt online to load changes, dynamic disturbance, and actuator noise.

Methods

  • Design: adaptive fuzzy PI controller (AFPIC) with a deep Q-learning network (DQLN) that dynamically adjusts the input/output scaling factors of the FPI controller; Q-network weights are updated using the Lyapunov stability theorem, and the target Q′-network is updated periodically.
  • Data/sample: practical implementation on an EHS160 electro-hydraulic servo system controlling hydraulic motor speed, using Arduino Mega for data acquisition and MATLAB for control; training used buffer B = 1000, 5000 episodes, max 1000 steps per episode, sampling time 10 ms, learning rate 0.001, discount factor 0.05, epsilon 0.9, and scaling factor ranges Ce [0.009, 0.1], Cde [0.005, 0.5], Cu [1, 9].
  • Main analysis: compared AFPIC-DQLN against PID, nonadaptive FPI, AFPIC-QL, and DACRL under step changes, 30% and 50% external loads, dynamic sinusoidal disturbance, and actuator noise; performance assessed with IAE, ISE, RMSE, transient response specifications, and computation time.

Limitations

  • The paper does not include a dedicated limitations section; it states future work will employ hierarchical deep reinforcement learning based on interval type-2 fuzzy systems to accommodate potential random perturbations and statistical experimental assessments (stated).
  • The proposed controller has higher computation time than all compared controllers (0.651 ms vs 0.045–0.597 ms), though the authors note it remains below the sampling period (stated in Remark 2).
  • Evident from the text: evaluation is confined to one EHS160 electro-hydraulic servo system and four practical tasks, and the Data availability statement says no datasets were generated or analyzed; training requires 5000 episodes with a 1000-sample buffer and 500-sample batch, which may limit generalizability.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1007/s00521-024-10741-x

Cites

42 of 42 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Image 1
From the PDF
Fig. 2 Structure of DQLN
Fig. 2 Structure of DQLNFrom the PDF
Fig. 3 Structure of Q-network in DQLN
Fig. 3 Structure of Q-network in DQLNFrom the PDF
Fig. 4 Structure of Q 0 -network in DQLN
Fig. 4 Structure of Q 0 -network in DQLNFrom the PDF
Fig. 5 Electro-hydraulic servo system with controller
Fig. 5 Electro-hydraulic servo system with controllerFrom the PDF
Fig. 6 Real system of the EHS160 module with controller
Fig. 6 Real system of the EHS160 module with controllerFrom the PDF
Fig. 7 Response of the EHS160 for task 1
Fig. 7 Response of the EHS160 for task 1From the PDF
Fig. 8 Control signal for task 1
Fig. 8 Control signal for task 1From the PDF
Fig. 9 Response of the EHS160 for task 2 (First scenario)
Fig. 9 Response of the EHS160 for task 2 (First scenario)From the PDF
Fig. 10 Control signal for task 2 (First scenario)
Fig. 10 Control signal for task 2 (First scenario)From the PDF
Fig. 11 Response of the EHS160 for task 3 (Second scenario)
Fig. 11 Response of the EHS160 for task 3 (Second scenario)From the PDF
Fig. 12 Control signal for task 3 (Second scenario)
Fig. 12 Control signal for task 3 (Second scenario)From the PDF
Fig. 13 Response of the EHS160 for task 3
Fig. 13 Response of the EHS160 for task 3From the PDF
Fig. 14 Control signal for task 3
Fig. 14 Control signal for task 3From the PDF
Fig. 15 Response of the EHS160 for task 4
Fig. 15 Response of the EHS160 for task 4From the PDF
Fig. 16 Control signal for task 4
Fig. 16 Control signal for task 4From the PDF

Verification

Claims the paper makes

1 Valid · 1 Not valid · 20 Not checked · 44 Own results and interpretations

27 more claims

Create a free account to see all 66 claims and to verify the 2 references still open.

See all claims — free account
  • It may state two separate claims as one. 20
  • Probably not a substantial claim. 17
  • The quote only partly states this claim. 1

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

2 of 2

Citations

Cites

Cited by

See all 17

Metadata

Authors
Khater, A. A., Fekry, M., El-Bardini, M., & El-Nagar, A. M.
Affiliations
Menoufia University (EG)
Corresponding author
El-Nagar, Ahmad M.
Venue
Neural Computing and Applications
Published
2025-01-10
Volume
37
Issue
30
Pages
24607-24624
ISSN
  • 0941-0643 (ISSN-L)
  • 1433-3058
OpenAlex
W4406237058
Cited by
17
References
36
Citations by year
2026: 10 · 2025: 7
FWCI
4.94
Type
journal article
Language
en
Open access
Licence
Locations
Neural Computing and Applications (publishedVersion, cc-by)
Funders
Minufiya University
Topics
  • Physical Sciences › Engineering › Mechanical Engineering › Hydraulic and Pneumatic Systems
  • Physical Sciences › Engineering › Control and Systems Engineering › Adaptive Control of Nonlinear Systems
  • Physical Sciences › Computer Science › Artificial Intelligence › Fuzzy Logic and Control Systems
Keywords
electro-hydraulic servo system, deep reinforcement learning, adaptive fuzzy control, fuzzy PI controller, Lyapunov stability, real-time parameter adaptation

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.